Method of augmenting the number of labeled images for training a neural network
View Patent ↗A method of augmenting the number of labeled images for training a neural network comprising the steps of—Starting from a dataset of labeled images with corresponding segmentation masks and a dataset of unlabeled images, gathering for a given image i in a data set of labeled images a number of images with similar metadata in said dataset of unlabeled images so as to form data sub-set Sim i,—Training a multiclass segmentation neural network on said labeled images thereby generating segmentation masks for the images in subset Sim i,—On the basis of these segmentation masks judging similarity between images of Sim i and image i and finding the most similar image(s) in Sim i by computing and comparing histograms of segmentation masks of image i and images in Sim i—Transferring the histogram of the most similar images in Sim i to given image i.
1. A method of augmenting the number of labeled images for training a neural network, the method comprising:
starting from a dataset of labeled images with corresponding segmentation masks and a dataset of unlabeled images,
gathering for a given image i in a data set of labeled images a number of images with metadata that have at least one item with the same value in said dataset of unlabeled images so as to form a data sub-set Sim i,
training a multiclass segmentation neural network on said labeled images thereby generating segmentation masks for the images in sub-set Sim i,
on the basis of these segmentation masks judging similarity between images of Sim i and image i and finding most similar image(s) in Sim i by computing histograms of segmentation masks of image i and images in Sim i and by comparing them, and
transferring the histogram of the most similar images in Sim i to given image i.
2. The method of claim 1 , wherein said image i and images in said sub- set Sim i are registered on top of each other before said histograms are compared.
3. The method of claim 2 , wherein as a postprocessing step an image is sorted out when use of such an image would render an overexposed or underexposed result.
4. The method of claim 1 , wherein as a postprocessing step an image is sorted out when use of such an image would render an overexposed or underexposed result.